anggakayy/xauusd-regime-switching-ensemble

Vektor & ScalpDaily Gold (XAUUSD) Scalper Ensemble

Kumpulan model Machine Learning (XGBoost Classifier/Regressor) berbasis data tabular/time-series untuk memprediksi arah pergerakan harga emas (XAUUSD) dengan presisi tinggi pada timeframe M5/M15.

Model ini secara dinamis berganti regime berdasarkan volatilitas pasar yang diukur menggunakan ADX (Average Directional Index):

1. Trend Engine (ADX > 25): Vektor v6 ML Model
2. Sideways Engine (ADX < 20): Vektor v8 Sideways Model
3. General Scalper (M15): ScalpDaily XGBoost Classifier

---

📊 Rangkuman Performa & Benchmark Model

Nama Model	Target Kondisi	Akurasi	Presisi (Win Rate BUY)	Recall	F1-Score	Data Latih (Size)
Vektor v6 ML (Trend)	Trending Market	44.84%	67.87%	27.38%	39.02%	33.248 Bars
Vektor v8 Sideways	Ranging Market	69.23%	75.00%	80.77%	77.78%	10.947 Bars
ScalpDaily XGBoost	M15 Scalping	71.43%	74.19%	85.19%	79.31%	1.034 Bars
Vektor v6 ML (Backup)	Backup/Optimized	59.97%	73.67%	68.11%	70.78%	10.304 Bars

---

📈 Visualisasi Grafik Performa Premium

Berikut adalah kompilasi visualisasi grafik kinerja model out-of-sample:

1. Vektor v8 Sideways Model (Regime Sideways)

Confusion Matrix	Feature Importance

ROC-AUC Curve	Equity Growth Curve

---

2. Vektor v6 ML Backup Model (Regime Trend - Optimized)

Confusion Matrix	Feature Importance

ROC-AUC Curve	Equity Growth Curve

---

🔍 Detail Spesifikasi Model

1. Vektor v8 Sideways Model (vektor_v8_sideways.joblib)

• Arsitektur: XGBoost (Optimasi Optuna Bayesian Hyperparameters)
• Hiperparameter Kunci: n_estimators=602, max_depth=3, learning_rate=0.0127, subsample=0.793
• Confusion Matrix: 6 True Negatives, 21 True Positives (Win Rate riil mencapai 75.00%).

2. Vektor v6 ML - Trend Model (vektor_v6_ml.joblib)

• Arsitektur: XGBoost Classifier
• Wawasan Data: Dilatih dengan 33.248 bar data M5 (~6 bulan rentang waktu).
• Karakteristik: Sangat ketat menyeleksi sinyal (Recall 27.38%) demi mengejar kestabilan presisi.

---

🛠️ Cara Menggunakan Model Secara Lokal

import joblib
import numpy as np

# Load model
model = joblib.load("vektor_v6_ml.joblib")

# Prediksi arah (gunakan data input 25 fitur sesuai metadata)
dummy_data = np.random.rand(1, 25)
buy_probability = model.predict_proba(dummy_data)[0][1]

print(f"Probabilitas arah BUY: {buy_probability * 100:.2f}%")